faster monster gen

src/lib/components/MonsterGenerator/MonsterGenerator.svelte

@@ -18,16 +18,15 @@
     userImage: null,
     imageCaption: null,
     monsterConcept: null,
+    monsterStats: null,
     imagePrompt: null,
     monsterImage: null,
     error: null,
     isProcessing: false
   });
   
-  // Prompt templates
-  const MONSTER_CONCEPT_PROMPT = (caption: string) => `Based on this image caption: "${caption}"
-
-Create a Pokémon-style monster that transforms the object into an imaginative creature. The monster should clearly be inspired by the object's appearance but reimagined as a living monster.
+  // Custom prompt for joy-caption-alpha-two to generate everything in one step
+  const MONSTER_GENERATION_PROMPT = `Based on this image create a Pokémon-style monster that transforms the object into an imaginative creature. The monster should clearly be inspired by the object's appearance but reimagined as a living monster.
 
 Guidelines:
 - Take the object's key visual elements (colors, shapes, materials) incorporating all of them into a single creature design
@@ -42,45 +41,23 @@ Include:
 - A creative name that hints at the original object
 - Physical description showing how the object becomes a creature
 - Special abilities derived from the object's function
-- Personality traits based on the object's purpose`;
+- Personality traits based on the object's purpose
+
+Format your response as:
+\`\`\`
+# {monster name}
+Object Rarity: {common, somewhat rare, very rare, extremely rare, legendary}
+## Monster Visual Description
+...
+## Monster Lore
+...
+\`\`\``;
 
   const IMAGE_GENERATION_PROMPT = (concept: string) => `Extract ONLY the visual appearance from this monster concept and describe it in one concise sentence:
 "${concept}"
 
 Focus on: colors, body shape, eyes, limbs, mouth, and key visual features. Omit backstory, abilities, and non-visual details.`;
   
-  const MONSTER_STATS_PROMPT = (concept: string) => `Convert the following monster concept into a JSON object with stats:
-"${concept}"
-
-The output should be formatted as a JSON instance that conforms to the JSON schema below.
-
-\`\`\`json
-{
-  "properties": {
-    "name": {"type": "string", "description": "The monster's unique name"},
-    "description": {"type": "string", "description": "A brief physical description of the monster's appearance"},
-    "rarity": {"type": "integer", "minimum": 0, "maximum": 100, "description": "How rare/unique the monster is (0=very common, 100=legendary)"},
-    "HP": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Health/vitality stat (0=fragile, 100=incredibly tanky)"},
-    "defence": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Defensive/armor stat (0=paper thin, 100=impenetrable fortress)"},
-    "attack": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Physical attack power (0=harmless, 100=devastating force)"},
-    "speed": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Movement and reaction speed (0=immobile, 100=lightning fast)"},
-    "specialPassiveTraitDescription": {"type": "string", "description": "Describe a passive ability that gives this monster a unique advantage in battle"},
-    "attackActionName": {"type": "string", "description": "Name of the monster's primary damage-dealing attack (e.g., 'Flame Burst', 'Toxic Bite')"},
-    "attackActionDescription": {"type": "string", "description": "Describe how this attack damages the opponent and any special effects"},
-    "buffActionName": {"type": "string", "description": "Name of the monster's self-enhancement ability (e.g., 'Iron Defense', 'Speed Boost')"},
-    "buffActionDescription": {"type": "string", "description": "Describe which stats are boosted and how this improves the monster's battle performance"},
-    "debuffActionName": {"type": "string", "description": "Name of the monster's enemy-weakening ability (e.g., 'Intimidate', 'Slow Poison')"},
-    "debuffActionDescription": {"type": "string", "description": "Describe which enemy stats are lowered and how this weakens the opponent"},
-    "specialActionName": {"type": "string", "description": "Name of the monster's ultimate move (one use per battle)"},
-    "specialActionDescription": {"type": "string", "description": "Describe this powerful finishing move and its dramatic effects in battle"}
-  },
-  "required": ["name", "description", "rarity", "HP", "defence", "attack", "speed", "specialPassiveTraitDescription", "attackActionName", "attackActionDescription", "buffActionName", "buffActionDescription", "debuffActionName", "debuffActionDescription", "specialActionName", "specialActionDescription"]
-}
-\`\`\`
-
-Remember to base the stats on how unique/powerful the original object was. Common objects should have lower stats, unique objects should have higher stats.
-
-Write your response within \`\`\`json\`\`\``;
 
   async function importPiclet(picletData: PicletInstance) {
     state.isProcessing = true;
@@ -132,26 +109,18 @@ Write your response within \`\`\`json\`\`\``;
     state.isProcessing = true;
     
     try {
-      // Step 1: Caption the image
+      // Step 1: Generate monster concept with joy-caption (includes lore, visual description, and rarity)
       await captionImage();
       await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
       
-      // Step 2: Generate monster concept
-      await generateMonsterConcept();
-      await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
-      
-      // Step 3: Generate monster stats
+      // Step 2: Generate monster stats based on the concept
       await generateStats();
       await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
       
-      // Step 4: Generate image prompt
-      await generateImagePrompt();
-      await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
-      
-      // Step 5: Generate monster image
+      // Step 3: Generate monster image
       await generateMonsterImage();
       
-      // Step 6: Auto-save the monster
+      // Step 4: Auto-save the monster
       await autoSaveMonster();
       
       state.currentStep = 'complete';
@@ -210,102 +179,69 @@ Write your response within \`\`\`json\`\`\``;
     try {
       const output = await joyCaptionClient.predict("/stream_chat", [
         state.userImage,
-        "Descriptive" as CaptionType,
+        "Creative" as CaptionType,
         "very long" as CaptionLength,
         [], // extra_options
         "", // name_input
-        ""  // custom_prompt
+        MONSTER_GENERATION_PROMPT  // custom_prompt
       ]);
       
       const [prompt, caption] = output.data;
+      // The caption now contains the full monster concept with lore, visual description, and rarity
       state.imageCaption = caption;
-      console.log('Caption generated:', caption);
+      state.monsterConcept = caption; // Store as concept since it's the full monster details
+      console.log('Monster concept generated:', caption);
     } catch (error) {
       handleAPIError(error);
     }
   }
   
-  async function generateMonsterConcept() {
-    state.currentStep = 'conceptualizing';
-    
-    if (!rwkvClient || !state.imageCaption) {
-      throw new Error('Text generation service not available or no caption');
-    }
-    
-    const conceptPrompt = MONSTER_CONCEPT_PROMPT(state.imageCaption);
-    const systemPrompt = "You are a creative Pokémon designer. Transform everyday objects into imaginative monsters by blending their characteristics with creature features. Think like the designers who turned a Pokéball into Voltorb, a keyring into Klefki, or an ice cream cone into Vanillite. Be creative and whimsical while keeping the object recognizable.";
-    
-    console.log('Generating monster concept with prompt:', conceptPrompt);
-    
-    try {
-      const output = await rwkvClient.predict("/chat", [
-        conceptPrompt,          // message
-        [],                     // chat_history
-        systemPrompt,          // system_prompt
-        2048,                  // max_new_tokens
-        0.7,                   // temperature
-        0.95,                  // top_p
-        50,                    // top_k
-        1.0                    // repetition_penalty
-      ]);
-      
-      console.log('Zephyr output:', output.data[0]);
-      let conceptText = output.data[0];
-      
-      state.monsterConcept = conceptText;
-      console.log('Monster concept generated:', state.monsterConcept);
-      
-      if (!state.monsterConcept || state.monsterConcept.trim() === '') {
-        throw new Error('Failed to generate monster concept - received empty response');
-      }
-    } catch (error) {
-      handleAPIError(error);
-    }
-  }
   
-  async function generateImagePrompt() {
-    state.currentStep = 'promptCrafting';
+  
+  async function generateMonsterImage() {
+    state.currentStep = 'generating';
     
-    if (!rwkvClient || !state.monsterConcept) {
-      throw new Error('Text generation service not available or no concept');
+    if (!fluxClient || !state.monsterConcept) {
+      throw new Error('Image generation service not available or no concept');
     }
     
-    const promptGenerationPrompt = IMAGE_GENERATION_PROMPT(state.monsterConcept);
-    const systemPrompt = "You are an expert at creating concise visual descriptions for image generation. Extract ONLY visual appearance details and describe them in ONE sentence (max 50 words). Focus on colors, shape, eyes, limbs, and distinctive features. Omit all non-visual information like abilities, personality, or backstory.";
+    // Extract the visual description from the joy-caption output
+    const visualDescMatch = state.monsterConcept.match(/## Monster Visual Description\s*\n([\s\S]*?)(?=##|$)/);
     
-    console.log('Generating image prompt from concept');
-    
-    try {
-      const output = await rwkvClient.predict("/chat", [
-        promptGenerationPrompt, // message
-        [],                     // chat_history
-        systemPrompt,          // system_prompt
-        1024,                   // max_new_tokens
-        0.7,                   // temperature
-        0.95,                  // top_p
-        50,                    // top_k
-        1.0                    // repetition_penalty
-      ]);
+    if (visualDescMatch && visualDescMatch[1]) {
+      state.imagePrompt = visualDescMatch[1].trim();
+      console.log('Extracted visual description for image generation:', state.imagePrompt);
+    } else {
+      // Fallback: use zephyr to extract visual description
+      if (!rwkvClient) {
+        throw new Error('Text generation service not available for fallback');
+      }
       
-      console.log('Image prompt output:', output);
-      let promptText = output.data[0];
+      const promptGenerationPrompt = IMAGE_GENERATION_PROMPT(state.monsterConcept);
+      const systemPrompt = "You are an expert at creating concise visual descriptions for image generation. Extract ONLY visual appearance details and describe them in ONE sentence (max 50 words). Focus on colors, shape, eyes, limbs, and distinctive features. Omit all non-visual information like abilities, personality, or backstory.";
       
-      state.imagePrompt = promptText;
-      console.log('Image prompt generated:', state.imagePrompt);
+      console.log('Falling back to zephyr for visual description extraction');
       
-      if (!state.imagePrompt || state.imagePrompt.trim() === '') {
-        throw new Error('Failed to generate image prompt - received empty response');
+      try {
+        const output = await rwkvClient.predict("/chat", [
+          promptGenerationPrompt, // message
+          [],                     // chat_history
+          systemPrompt,          // system_prompt
+          1024,                   // max_new_tokens
+          0.7,                   // temperature
+          0.95,                  // top_p
+          50,                    // top_k
+          1.0                    // repetition_penalty
+        ]);
+        
+        state.imagePrompt = output.data[0];
+      } catch (error) {
+        handleAPIError(error);
       }
-    } catch (error) {
-      handleAPIError(error);
     }
-  }
-  
-  async function generateMonsterImage() {
-    state.currentStep = 'generating';
     
-    if (!fluxClient || !state.imagePrompt) {
-      throw new Error('Image generation service not available or no prompt');
+    if (!state.imagePrompt || state.imagePrompt.trim() === '') {
+      throw new Error('Failed to extract visual description');
     }
     
     try {
@@ -361,7 +297,61 @@ Write your response within \`\`\`json\`\`\``;
       throw new Error('Text generation service not available or no concept');
     }
     
-    const statsPrompt = MONSTER_STATS_PROMPT(state.monsterConcept);
+    // Extract rarity from the joy-caption output
+    let rarityScore = 50; // default
+    const rarityMatch = state.monsterConcept.match(/Object Rarity:\s*(common|somewhat rare|very rare|extremely rare|legendary)/i);
+    if (rarityMatch) {
+      const rarityMap: { [key: string]: number } = {
+        'common': 20,
+        'somewhat rare': 40,
+        'very rare': 60,
+        'extremely rare': 80,
+        'legendary': 100
+      };
+      rarityScore = rarityMap[rarityMatch[1].toLowerCase()] || 50;
+    }
+    
+    // Extract monster name from the concept
+    const nameMatch = state.monsterConcept.match(/^#\s+(.+)$/m);
+    const monsterName = nameMatch ? nameMatch[1] : 'Unknown Monster';
+    
+    // Update stats prompt to include the rarity score
+    const statsPrompt = `Convert the following monster concept into a JSON object with stats:
+"${state.monsterConcept}"
+
+The monster's rarity has been determined as: ${rarityScore}/100
+The monster's name is: ${monsterName}
+
+The output should be formatted as a JSON instance that conforms to the JSON schema below.
+
+\`\`\`json
+{
+  "properties": {
+    "name": {"type": "string", "description": "The monster's unique name"},
+    "description": {"type": "string", "description": "A brief physical description of the monster's appearance"},
+    "rarity": {"type": "integer", "minimum": 0, "maximum": 100, "description": "How rare/unique the monster is (0=very common, 100=legendary)"},
+    "HP": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Health/vitality stat (0=fragile, 100=incredibly tanky)"},
+    "defence": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Defensive/armor stat (0=paper thin, 100=impenetrable fortress)"},
+    "attack": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Physical attack power (0=harmless, 100=devastating force)"},
+    "speed": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Movement and reaction speed (0=immobile, 100=lightning fast)"},
+    "specialPassiveTraitDescription": {"type": "string", "description": "Describe a passive ability that gives this monster a unique advantage in battle"},
+    "attackActionName": {"type": "string", "description": "Name of the monster's primary damage-dealing attack (e.g., 'Flame Burst', 'Toxic Bite')"},
+    "attackActionDescription": {"type": "string", "description": "Describe how this attack damages the opponent and any special effects"},
+    "buffActionName": {"type": "string", "description": "Name of the monster's self-enhancement ability (e.g., 'Iron Defense', 'Speed Boost')"},
+    "buffActionDescription": {"type": "string", "description": "Describe which stats are boosted and how this improves the monster's battle performance"},
+    "debuffActionName": {"type": "string", "description": "Name of the monster's enemy-weakening ability (e.g., 'Intimidate', 'Slow Poison')"},
+    "debuffActionDescription": {"type": "string", "description": "Describe which enemy stats are lowered and how this weakens the opponent"},
+    "specialActionName": {"type": "string", "description": "Name of the monster's ultimate move (one use per battle)"},
+    "specialActionDescription": {"type": "string", "description": "Describe this powerful finishing move and its dramatic effects in battle"}
+  },
+  "required": ["name", "description", "rarity", "HP", "defence", "attack", "speed", "specialPassiveTraitDescription", "attackActionName", "attackActionDescription", "buffActionName", "buffActionDescription", "debuffActionName", "debuffActionDescription", "specialActionName", "specialActionDescription"]
+}
+\`\`\`
+
+Remember to set the rarity to ${rarityScore} and the name to "${monsterName}". Base the other stats on this rarity level.
+
+Write your response within \`\`\`json\`\`\``;
+    
     const systemPrompt = "You are a game designer specializing in monster stats and abilities. You must ONLY output valid JSON that matches the provided schema exactly. Do not include any text before or after the JSON. Do not include null values in your JSON response. Your entire response should be wrapped in a ```json``` code block.";
     
     console.log('Generating monster stats from concept');
@@ -542,13 +532,9 @@ Write your response within \`\`\`json\`\`\``;
       <div class="spinner"></div>
       <p class="processing-text">
         {#if state.currentStep === 'captioning'}
-          Analyzing your image...
-        {:else if state.currentStep === 'conceptualizing'}
-          Creating monster concept...
+          Creating monster concept from your image...
         {:else if state.currentStep === 'statsGenerating'}
           Generating battle stats...
-        {:else if state.currentStep === 'promptCrafting'}
-          Crafting generation prompt...
         {:else if state.currentStep === 'generating'}
           Generating your monster...
         {/if}

src/lib/components/MonsterGenerator/WorkflowProgress.svelte

@@ -22,28 +22,18 @@
     },
     {
       id: 'captioning',
-      label: 'Analyzing',
-      description: 'Creating detailed description'
-    },
-    {
-      id: 'conceptualizing',
-      label: 'Conceptualizing',
-      description: 'Designing your monster'
+      label: 'Monster Design',
+      description: 'Creating concept & lore'
     },
     {
       id: 'statsGenerating',
-      label: 'Generating Stats',
-      description: 'Creating battle attributes'
-    },
-    {
-      id: 'promptCrafting',
-      label: 'Crafting Prompt',
-      description: 'Preparing image generation'
+      label: 'Battle Stats',
+      description: 'Generating abilities'
     },
     {
       id: 'generating',
-      label: 'Generating',
-      description: 'Creating monster image'
+      label: 'Image Generation',
+      description: 'Creating monster art'
     },
     {
       id: 'complete',